`Geographically Diverse Commercial Properties
`Using Artificial Neural Networks
`
`Richard A. Borst, President
`Applied Image & Information Division
`Day & Zimmermann, Inc.
`
`Copyright © 1993 by Richard A. Botet
`(Authors Note: At the time of submission, the data conversion portion äf the project has not
`been completed. Therefore, statistical results were not available. TIey will be presented at the
`Fall 1993 IAAO Conference, and a written paper will also be available at that time.)
`
`OUTLINE OF PAPER
`
`Introduction
`The appraisal problem is defined. A New England based financial institution holds a
`portfolio of some 3,500 commercial properties with values in excess of one million dollars each.
`By agreement with Federal regulators, each property in this portfolio must have its appraisal
`updated annually. Given that the average cost of an appraisal update is approximately $3,500,
`an annual cost of over twelve million dollars is being incurred to meet the requirement.
`
`The Solution Approach
`Basically the solution approach is to apply the techniques of computer assisted mass
`appraisal (CAMA) to the problem to determine if defensible estimates of value can be produced,
`on some high percentage of the properties. With reliable value estimates, the case will be made
`to regulators for an alternative mechanism to establishing the values within the real estate
`portfolio.
`
`Two model formulations were considered. The first relied on the existence of properties
`which had been appraised more than once, and at different points in time. This allowed
`construction of a model that would allow for estimation of appraised value given a previous
`appraised value. The second model formulation was the more typical approach of estimating
`appraised value (not selling price) from property characteristics and other economic factors.
`
`2E?
`
`
`
`Tools Used
`A number of software products were used in performing the analysis including:
`
`ANALYSIS TOOLS
`
`Tool
`
`Use
`
`Paradox for Windows
`
`Database Construction/Analysis Reporting
`
`Map Info for Windows
`
`Locational Analysis
`
`Brain Maker
`
`Neural Network Under Model Building arid Calibration
`
`Number Cruncher
`Statistical System
`
`Multivariate Analysis
`
`Results and Conclusions
`
`Forthcoming at Conference.
`
`258
`
`